Reservoir Characterization from 3-D Seismic Data Using Artificial Neural Networks and Stochastic Modeling Techniques

For an accurate prediction of the production profile of a hydrocarbon reservoir, an optimal assessment of the geological reservoir model is required. The reservoir architecture and lithostratigraphic properties of the model are the most important boundary conditions in the economic evaluation of the reservoir. In this paper, I will show how seismic data can be used to extract lithostratigraphic information to constrain the reservoir model. Two innovative seismic inversion schemes based on the application of artificial neural networks are presented to achieve this goal. Method 1 is a deterministic approach; [open quotes]back-propagation[close quotes] networks are trained by offering seismic responses at well location as input nodes and well results, e.g., reservoir porosity and/or net-to-gross ratio's, as output nodes. This method can be applied on existing fields with sufficient well control only. Method 2 is a stochastic approach that can be employed in areas with limited well control. Synthetic seismograms are created by stochastically varying the model input parameters such as layer thicknesses, densities, and velocities. The networks are trained by offering the filtered synthetic seismograms an input nodes and one (or more) of the underlying model parameters as the output nodes. In both methods, the trained networks are tested onmore » independent data sets to obtain a measure for the accuracy of the obtained results. The trained and tested networks are subsequently applied to the real seismic data. The techniques discussed in this paper are to be implemented in an industrial quality software package by the [open quotes]Probe[close quotes] consortium.« less